Overview

This project will examine the factors related to the length of stay for pediatric patients with behavioral health concerns in the CHOP emergency department.

Staff involved on a consultation basis are ED Psychiatrist Andi Fu, ED Physician Jeremy Espositio, and ED Physician and Director of Emergency Information Systems, Joseph Zorc, MD.

Introduction

According to national surveys, as many as one in six U.S. children between the ages of 6 and 17 has a treatable mental health disorder (Whitney and Peterson 2019). For many parents of children struggling with behavioral health needs, taking their child to the emergency department (ED) may be their first contact and last resort to access behavioral health services. However, emergency departments are not well suited to meet behavioral service needs. The increased need for behavioral health services nationally is outpacing the rate of available behavioral health providers. As such, children who present to the ED with behavioral health needs are susceptible to prolonged length of stays (LOS) with some studies showing longer LOS than patients without behavioral health conditions(Case et al. 2011).

Several studies have examined the relationship between patient and hospital characteristics and LOS, but to date, none have examined the additional influence of within-ED behavioral health actions and decisions on LOS (Chakravarthy et al. 2017; Nash et al. 2021; Case et al. 2011).

This project will include demographic, emergency department, and behavioral health activity factors. Identifying and understanding the relative impact of these factors may help with triaging and disposition decisions to reduce LOS. This reduction may not necessarily lead to a more immediate implementation of behavioral health services, but it could still improve patient satisfaction, ED throughput, and lower costs.

This project is interdisciplinary due to the collaboration between medical ED providers and staff and behavioral health providers. Patients with behavioral health conditions are present everywhere within a hospital, but their main points of entry are: directly with a behavioral health department or indirectly through primary care and the emergency department. For the latter, this means such patients are requesting assistance from providers who are not specifically trained in assessing or managing patients with behavioral health conditions. On the other hand, behavioral health providers may be less likely to have experience working with patients with high acuity or co-morbid medical conditions. My discussions with colleagues have provided me with information about the range of factors that need to be considered in determining what in hospital and outside of the hospital services are used to help patients during their admission. CHOP has an outdated dashboard with a few relevant factors, but it is outdated and has not incorporated recent developments in tracking data about these patients.

Goals

  • To examine demographic and clinical risk factors associated with length of stay (LOS) and prolonged LOS (>= 24 hrs).
  • To illustrate dispositions recommended by the CHOP behavioral health and social work teams.

Methods

The cohort for this analysis is patients with mental health emergencies seen in the CHOP Emergency Department (ED and EDECU - Emergency Department Extended Care Unit). Data were queried from CHOP’s data warehouse (CDW), which is built from data stored in the EPIC EHR.

  • Admissions were included in this analysis based on:

    • Admissions occurring from 2020 to the present.

    • Patients being between 5 and 18 years old at the time of the ED admission as mental health diagnoses are uncommon in infants and toddlers.

    • Having at least one of the several Inclusion Variables focusing on ED care activities and behavioral-health data obtained during the admission.

Variables with many levels were grouped to the top 5 to make comparisons easier to interpret.

Primary outcomes were the total length of stay (LOS) as a continuous variable and prolonged length of stay as a categorical variable (< 24 hours versus >=24 hours).

Results

Data Analysis

Descriptive statistics and plots were created to understand the nature of LOS in this cohort.

To identify risk factors associated with prolonged LOS, we conducted logistic regression analyses, calculating odds ratios (OR) with 95% confidence intervals (CI). For the multivariate regression model, we selected the same variables.

## Get data from the CDW
ed_encs_raw <- run_sql("select * from marts_dev.proctors.bh_ed_encs", dsn = "CDWUAT", lowercase_names = TRUE)

## Start data prep by re-coding indicators and lumping factors with values that are low in frequency.
df <- ed_encs_raw %>%
  # filter(age_at_visit <= 18) %>%
  mutate(weekday_ind = as.numeric(weekday_ind)) %>%
  mutate_at(
    vars(ends_with("ind") & !starts_with("mychop")),
    ~ (recode_factor(., "1" = "1", "0" = "0", .missing = "0"))
  ) %>% # take all of the indicators and make the
  # missing zeros. Due to how this was not set within the SQL
  mutate_at(
    vars(ends_with("ind") & !starts_with("mychop")),
    ~ (fct_inseq(.))
  ) %>%
  mutate( ## create and modify factors including cleaning up text values
    mychop_activation_ind = recode_factor(mychop_activation_ind, "1" = "Active", "0" = "Inactive"),
    sex = recode_factor(sex, M = "Male", F = "Female", U = "Unknown"),
    payor_group = as.factor(str_to_title(payor_group)),
    primary_medical_diagnosis_name = case_when(
      str_detect(primary_medical_diagnosis_name, regex("poison", ignore_case = T)) ~ "Overdose",
      TRUE ~ primary_medical_diagnosis_name
    ),
    ed_discharge_disposition =
      case_when(
        str_detect(ed_discharge_disposition, "Psychiatric") ~ "Psychiatry Facility",
        TRUE ~ ed_discharge_disposition
      ),
    ed_los_hrs_cat = if_else(total_ed_hrs >= 24, "ED LOS >= 24", "ED LOS < 24"),
    ed_los_hrs_cat_12 = if_else(total_ed_hrs >= 12, "ED LOS >= 12", "ED LOS < 12"),
    ed_discharge_disposition = fct_lump_n(ed_discharge_disposition, 10),
    preferred_language = fct_lump_n(preferred_language, 2),
    race = fct_lump_n(race, 5),
    primary_bh_diagnosis_name = fct_lump_n(primary_bh_diagnosis_name, 10),
    primary_medical_diagnosis_name = fct_lump_n(primary_medical_diagnosis_name, 5),
    ed_discharge_destination = fct_lump_n(ed_discharge_destination, 5),
    ed_arrival_mode = fct_lump_n(ed_arrival_mode, 5),
    ed_psych_discharge_location = fct_lump(ed_psych_discharge_location, 10),
    race_ethnicity = fct_lump(race_ethnicity, 5),
    mailing_zip = fct_lump(mailing_zip, 5),
    county = fct_lump(county, 5),
    bhs_suicidal_ideation_score = case_when(
      str_detect(bhs_suicidal_ideation_score, "and no current or past suicidal thoughts") ~ "No past or current SI",
      str_detect(bhs_suicidal_ideation_score, "and no current suicidal thoughts") ~ "Lifetime SI",
      str_detect(bhs_suicidal_ideation_score, "and current suicidal thoughts") ~ "Current SI",
      TRUE ~ bhs_suicidal_ideation_score
    ),
    edecu_los_hrs = na_if(edecu_los_hrs, 0) ## remove zeros as they are not measured as a zero, just not existing.
  ) %>%
  mutate( # changing the levels on factors
    race = fct_relevel(race, "White", "Black or African American", "Asian"),
    payor_group = fct_relevel(payor_group, "Medical Assistance"),
    bhs_depression_score = fct_relevel(bhs_depression_score, "Mild Depression", "Moderate Depression", "Severe Depression"),
    ed_discharge_disposition = fct_relevel(ed_discharge_disposition, "Discharge"),
    bh_discharge_disposition = fct_relevel(bh_discharge_disposition, "No Intervention"),
    bhs_suicidal_ideation_score = fct_relevel(bhs_suicidal_ideation_score, "No past or current SI", "Current SI"),
    ethnicity = fct_relevel(ethnicity, "Not Hispanic Or Latino"),
    arrival_shift = fct_relevel(arrival_shift, "Day (8a-4p)"),
    week_day = fct_relevel(week_day, "Mon", "Tue", "Wed", "Thu", "Fri", "Sat", "Sun"),
    age_group = fct_relevel(age_group, "5-8", "9-12", "13-15", "16-18"),
    race_ethnicity = fct_relevel(race_ethnicity, "Non-Hispanic White"),
    bhs_depression_score = fct_explicit_na(bhs_depression_score, na_level = "(Not Assessed)"),
    bhs_suicidal_ideation_score = fct_explicit_na(bhs_suicidal_ideation_score, na_level = "(Not Assessed)"),
    primary_medical_diagnosis_name = fct_explicit_na(primary_medical_diagnosis_name, na_level = "(BH diagnosis)"),
    primary_bh_diagnosis_name = fct_explicit_na(primary_bh_diagnosis_name, na_level = "(Medical diagnosis)"),
    primary_bh_diagnosis_name = fct_relevel(primary_bh_diagnosis_name, "(Medical diagnosis)"),
    primary_medical_diagnosis_name = fct_relevel(primary_medical_diagnosis_name, "(BH diagnosis)"),
    ed_discharge_destination = fct_relevel(ed_discharge_destination, "Patient/family/home"),
    county = fct_relevel(county, "Philadelphia")
  )

## Name columns that we will turn to factors
cols <- c(
  "pediatric_age_days_group", "sex", "race", "ethnicity", "payor_group", "race_ethnicity", "primary_diagnosis_category", "ed_los_hrs_cat", "bhs_depression_score",
  "bhs_suicidal_ideation_score", "ed_discharge_disposition", "preferred_language",
  "ed_discharge_location", "ed_psych_discharge_location", "bh_discharge_disposition", "arrival_shift", "primary_bh_diagnosis_name", "primary_medical_diagnosis_name", "age_group", "week_day", "ed_arrival_mode", "ed_discharge_destination"
)

df[cols] <- lapply(df[cols], as.factor)

df <- df %>%
  set_variable_labels(.labels = nice_display_names(., keep_uppercase = c("bh", "edecu", "bhs", "bhip", "mh", "ed", "csn", "mrn", "ip", "zip", "asq"))) # set labels that will make text in charts look formatted better

Length of Stay Analysis

Charting the length of stay over time using a statistical process control chart (SPC) chart. This will identify areas where the data are well above or below the established median during the length of time in the chart.

ED LOS and EDECU LOS are calculated based on the difference between their arrival time and their departure time. A total LOS was calculated as the sum of both LOS values.

# Chart the Median LOS for just ED LOS
df %>%
  remove_incomplete_end_dates(ed_admit_date, period = "month") %>%
  hc_spc(
    # data = df,
    x = ed_admit_date,
    y = ed_los_hrs, chart = "xbar",
    title = "Median ED Length of Stay", xlab = "Month Year", ylab = "Hours", agg.fun = "median"
  )
# Chart the Median LOS for just EDECU LOS
df %>%
  remove_incomplete_end_dates(ed_admit_date, period = "month") %>%
  hc_spc(
    # data = df,
    x = ed_admit_date,
    y = edecu_los_hrs, chart = "xbar",
    title = "Median EDECU Length of Stay", xlab = "Month Year", ylab = "Hours",
    agg.fun = "median"
  )
# Chart the Median LOS for the sum of ED and EDECU LOS
df %>%
  remove_incomplete_end_dates(ed_admit_date, period = "month") %>%
  hc_spc(
    # data = df,
    x = ed_admit_date,
    y = total_ed_hrs, chart = "xbar",
    title = "Median Total (ED & EDECU) Length of Stay", xlab = "Month Year", ylab = "Hours",
    agg.fun = "median"
  )

Distribution of LOS across the three measures. They are highly skewed, which is typical for LOS measurements across studies. LOS can be transformed to be a more normal distribution for statistical analysis, but that was not done in this analysis as transformation can make interpreting the results less straightforward.

# Create historgrams for all three measures of LOS

los_ed_hrs_histogram <-
  ggplot(df) +
  ggtitle("ED LOS") +
  geom_histogram(aes(x = ed_los_hrs), fill = chop_colors("blue")) +
  labs(
    x = "Hours",
    y = "Total Admissions",
    title = "ED LOS"
  ) +
  theme_chop()

los_ededu_hrs_histogram <-
  ggplot(df) +
  ggtitle("EDECU LOS") +
  geom_histogram(aes(x = edecu_los_hrs), fill = chop_colors("blue")) +
  labs(
    x = "Hours",
    y = "Total Admissions",
    title = "EDECU LOS"
  ) +
  theme_chop()

los_total_hrs_histogram <-
  ggplot(df) +
  ggtitle("Total LOS") +
  geom_histogram(aes(x = total_ed_hrs),
    fill = chop_colors("blue"),
    binwidth = 5
  ) +
  labs(
    x = "Hours",
    y = "Total Admissions",
    title = "Total LOS"
  ) +
  theme_chop()

## Arrange them as a grid.
grid.arrange(los_ed_hrs_histogram, los_ededu_hrs_histogram, los_total_hrs_histogram,
  widths = c(0.3, 0.3, 0.3), nrow = 1
)

Cohort Data Sample Overview

Patient Characteristics

Demographic and other patient-level characteristics.

Summary Highlights: Most patients are females above the age of 13, and are non-hispanic blacks. In addition, most patients are from Philadelphia, are primarily English speaking and have Medical Assistance.

## Create an easy to re-use column grouping
pat_chars <- c("age_group", "sex", "race_ethnicity", "county", "preferred_language", "payor_group", "mychop_activation_ind")

df_char_summary <- df %>%
  select(all_of(pat_chars), total_ed_hrs)

## Create good looking summary tables
tbl1 <-
  df_char_summary %>%
  mutate_at(
    vars(ends_with("ind")),
    ~ (recode(., "1" = "Yes", "0" = "No"))
  ) %>%
  tbl_summary(
    include = pat_chars,
    statistic = list(
      all_continuous() ~ "{mean} ({sd})" # ,
    ),
    digits = all_continuous() ~ 2,
    missing = "no"
  ) %>%
  add_n() %>%
  modify_caption("**Patient Characteristics**") %>%
  modify_header(label ~ "**Variable**")

tbl2 <-
  df_char_summary %>%
  mutate_at(
    vars(ends_with("ind")),
    ~ (recode(., "1" = "Yes", "0" = "No"))
  ) %>%
  tbl_continuous(
    variable = total_ed_hrs,
    include = pat_chars
  ) %>%
  modify_header(all_stat_cols() ~ "**Median LOS in Hours**")

tbl_final <-
  tbl_merge(list(tbl1, tbl2)) %>%
  modify_spanning_header(everything() ~ NA)

tbl_final
Patient Characteristics
Variable N N = 7,9191 Median LOS in Hours2
Age Group 7,919
5-8 826 (10%) 5 (3, 7)
9-12 1,756 (22%) 5 (4, 9)
13-15 3,056 (39%) 6 (4, 10)
16-18 2,281 (29%) 6 (4, 9)
Sex 7,919
Male 2,859 (36%) 5 (4, 8)
Female 5,059 (64%) 6 (4, 9)
Unknown 1 (<0.1%) 8 (8, 8)
Race Ethnicity 7,918
Non-Hispanic White 2,856 (36%) 6 (4, 8)
Asian 236 (3.0%) 5 (4, 8)
Hispanic or Latino 713 (9.0%) 6 (4, 8)
Non-Hispanic Black 3,473 (44%) 5 (4, 9)
Other 640 (8.1%) 6 (4, 9)
Unknown 1
County 7,917
Philadelphia 3,853 (49%) 5 (4, 9)
Bucks 446 (5.6%) 6 (5, 9)
Chester 257 (3.2%) 6 (4, 9)
Delaware 1,125 (14%) 5 (4, 9)
Montgomery 845 (11%) 6 (4, 9)
Other 1,391 (18%) 6 (4, 8)
Unknown 2
Preferred Language 7,901
English 7,652 (97%) 6 (4, 9)
Spanish 138 (1.7%) 6 (4, 7)
Other 111 (1.4%) 6 (5, 9)
Unknown 18
Payor Group 7,798
Medical Assistance 4,354 (56%) 6 (4, 9)
Charity Care 1 (<0.1%) 2 (2, 2)
Commercial 3,297 (42%) 6 (4, 8)
Other 146 (1.9%) 5 (4, 8)
Unknown 121
Mychop Activation 7,724
Active 5,472 (71%) 6 (4, 9)
Inactive 2,252 (29%) 6 (4, 9)
Unknown 195

1 n (%)

2 Median (IQR)

ED Characteristics

These include factors determined at the beginning or end of the admission.

Summary Highlights: Most patients arrive during the day, during the school months, and via cars. The most prevalent chief complaint is “Psych Emergency” or other chief complaints with mental health types of labels. The Majority of patients are discharged home from the ED. The most common inpatient Psychiatry Discharge setting is Fairmount, followed by Belmont.

## Create an easy to re-use column grouping
ed_chars <- c('arrival_shift', 'week_day', 'school_month_ind', 'ed_arrival_mode','psych_emergency_cc_ind', 'altered_mental_status_cc_ind', 'mental_health_cc_ind', 'hr_72_revisit_ind', 'ed_discharge_disposition', 'ed_psych_discharge_location', 'ed_discharge_location', 'ed_discharge_destination')

## Create good looking summary tables
df_ed_chars_summary <- df %>% 
  select(all_of(ed_chars), total_ed_hrs)

tbl3 <-
df_ed_chars_summary %>% 
    mutate_at(
    vars(ends_with("ind")),
    ~ (recode(., "1" = "Yes", "0" = "No"))
  ) %>%
  tbl_summary(
    include =  all_of(ed_chars),
        statistic = list(all_continuous() ~ "{mean} ({sd})"
                     ),
    digits = all_continuous() ~ 2
    , missing = "no"
       ) %>% 
  add_n() %>% 
  modify_caption("**Patient Characteristics**")  %>%
  modify_header(label ~ "**Variable**")

tbl4 <- 
   df_ed_chars_summary %>% 
    mutate_at(
    vars(ends_with("ind")),
    ~ (recode(., "1" = "Yes", "0" = "No"))
  ) %>%
  tbl_continuous(
    variable = total_ed_hrs,
    include =  all_of(ed_chars)
  ) %>%
  modify_header(all_stat_cols() ~ "**Median LOS in Hours**")

tbl_final2 <-
  tbl_merge(list(tbl3, tbl4)) %>%
  modify_spanning_header(everything() ~ NA)

tbl_final2
Patient Characteristics
Variable N N = 7,9191 Median LOS in Hours2
Arrival Shift 7,919
Day (8a-4p) 4,135 (52%) 6 (4, 8)
Evening (4p-12a) 3,089 (39%) 5 (4, 8)
Overnight (12a-8a) 695 (8.8%) 7 (4, 13)
Week Day 7,919
Mon 1,295 (16%) 6 (4, 9)
Tue 1,343 (17%) 6 (4, 9)
Wed 1,305 (16%) 6 (4, 9)
Thu 1,262 (16%) 6 (4, 9)
Fri 1,117 (14%) 6 (4, 8)
Sat 765 (9.7%) 6 (4, 9)
Sun 832 (11%) 5 (4, 8)
School Month 7,919 6,735 (85%)
No 6 (4, 10)
Yes 6 (4, 8)
ED Arrival Mode 7,919
Car 3,372 (43%) 6 (4, 9)
Ems-other 86 (1.1%) 7 (5, 11)
Not Applicable 2,737 (35%) 6 (4, 9)
Private Vehicle 1,258 (16%) 6 (4, 8)
Other 466 (5.9%) 5 (4, 8)
Psych Emergency Cc 7,919 3,548 (45%)
No 5 (4, 8)
Yes 6 (4, 18)
Altered Mental Status Cc 7,919 167 (2.1%)
No 6 (4, 9)
Yes 6 (5, 9)
Mental Health Cc 7,919 4,248 (54%)
No 6 (4, 8)
Yes 6 (4, 13)
Hr 72 Revisit 7,919 175 (2.2%)
No 6 (4, 9)
Yes 6 (4, 9)
ED Discharge Disposition 7,919
Discharge 4,286 (54%) 5 (3, 6)
Admit 3,011 (38%) 7 (5, 9)
AMA 6 (<0.1%) 5 (3, 5)
EDECU 9 (0.1%) 5 (4, 8)
Left Before Discharge (Eloped) 14 (0.2%) 5 (4, 6)
LWBS After Triage 4 (<0.1%) 3 (3, 3)
Non-Patient ED Screen with Refusal of Treatment/Transfer 1 (<0.1%) 26 (26, 26)
NOT APPLICABLE 6 (<0.1%) 6 (6, 7)
Psychiatry Facility 553 (7.0%) 41 (24, 86)
Transfer from Triage (Other) 2 (<0.1%) 119 (89, 148)
Transfer from Triage to HUP 2 (<0.1%) 8 (7, 10)
Transfered to Another Facility(Not from Triage) 25 (0.3%) 18 (7, 60)
ED Psych Discharge Location 493
Belmont 111 (23%) 30 (20, 71)
Devereux 9 (1.8%) 41 (23, 258)
Fairmount 163 (33%) 43 (25, 93)
Foundations 22 (4.5%) 38 (20, 85)
Friends 54 (11%) 34 (20, 64)
Horsham Clinic 116 (24%) 42 (25, 88)
Kids Clinic 11 (2.2%) 57 (49, 96)
Rockford 7 (1.4%) 49 (31, 64)
Unknown 7,426
ED Discharge Location 7,919
ED 3,960 (50%) 4 (3, 6)
EDECU 1,080 (14%) 46 (23, 98)
Inpatient 2,521 (32%) 6 (5, 8)
MBU 358 (4.5%) 7 (5, 9)
ED Discharge Destination 7,919
Patient/family/home 1,212 (15%) 7 (5, 9)
Not Applicable 6,337 (80%) 5 (4, 8)
Transfer To Another Facility 78 (1.0%) 7 (5, 11)
Transfer To Psych Facility 209 (2.6%) 8 (6, 14)
Transfer To Rehab-cssh 29 (0.4%) 5 (3, 7)
Other 54 (0.7%) 7 (5, 10)

1 n (%)

2 Median (IQR)

#footnote = AMA= against medical advice;

Diagnosis Characteristics

These include diagnosis groups for behavioral health and non-behavioral health conditions.

Summary Highlights: The majority of patients have a primary diagnoses involving a behavioral health condition. Among the behavioral health diagnoses, suicidal ideation/self-injury are the most prevalent, followed by aggression, agitation, and anger, eating disorders, and substance use disorders. Most patients do not have a complex chronic condition and are not medically complex.

## Create an easy to re-use column grouping
dx_chars <- c('primary_diagnosis_is_bh_ind', 'primary_bh_diagnosis_name', 'primary_medical_diagnosis_name', 'complex_chronic_condition_ind', 'medically_complex_ind')

df_dx_chars_summary <- df %>% 
  select(all_of(dx_chars), total_ed_hrs)

## Create good looking summary tables
tbl5 <- 
df_dx_chars_summary %>% 
    mutate_at(
    vars(ends_with("ind")),
    ~ (recode(., "1" = "Yes", "0" = "No"))
  ) %>%
  tbl_summary(
    include =  all_of(dx_chars),
    statistic = list(all_continuous() ~ "{mean} ({sd})"
                     ),
    digits = all_continuous() ~ 2,
    missing = "no",
    missing_text = "(Missing)"
              ) %>% 
  modify_caption("**Diagnosis Characteristics**")  %>%
  modify_header(label ~ "**Variable**")


tbl6 <- 
   df_dx_chars_summary %>% 
      mutate_at(
    vars(ends_with("ind")),
    ~ (recode(., "1" = "Yes", "0" = "No"))
  ) %>%
  tbl_continuous(
    variable = total_ed_hrs,
    include =  all_of(dx_chars)
  ) %>%
  modify_header(all_stat_cols() ~ "**Median LOS in Hours**")

tbl_final3 <-
  tbl_merge(list(tbl5, tbl6)) %>%
  modify_spanning_header(everything() ~ NA)

tbl_final3
Diagnosis Characteristics
Variable N = 7,9191 Median LOS in Hours2
Primary Diagnosis Is BH 4,485 (57%)
No 6 (4, 8)
Yes 6 (4, 10)
Primary BH Diagnosis Name
(Medical diagnosis) 3,434 (43%) 6 (4, 8)
Aggression, agitation, anger 1,113 (14%) 5 (4, 10)
Altered mental status 383 (4.8%) 5 (4, 7)
Anxiety disorders 392 (5.0%) 4 (3, 5)
Eating disorders 767 (9.7%) 6 (5, 7)
Hallucinations 60 (0.8%) 6 (4, 9)
Other mental health condition 13 (0.2%) 4 (2, 6)
Somatic disorders 53 (0.7%) 5 (3, 7)
Substances use disorders 506 (6.4%) 6 (4, 9)
Suicidal ideation, self-injury 1,198 (15%) 9 (5, 50)
Primary Medical Diagnosis Name
(BH diagnosis) 5,287 (67%) 5 (4, 9)
Abdominal And Pelvic Pain 159 (2.0%) 7 (5, 9)
Convulsions, Not Elsewhere Classified 91 (1.1%) 6 (4, 8)
Nausea And Vomiting 81 (1.0%) 7 (6, 9)
Overdose 172 (2.2%) 7 (5, 14)
Pain In Throat And Chest 88 (1.1%) 5 (4, 6)
Other 2,041 (26%) 6 (4, 8)
Complex Chronic Condition 1,849 (23%)
No 6 (4, 9)
Yes 6 (4, 8)
Medically Complex 861 (11%)
No 6 (4, 9)
Yes 6 (4, 8)

1 n (%)

2 Median (IQR)

Behavioral Health Activity Characteristics

A summary of the behavioral health-related activities that were documented during a patient’s admission. Includes mental health screenings, notes documented, medications, safety orders, and clinical dispositions.

Summary Highlights: For behavioral health assessments, severe depression was the most common score on the BHS for those who had a history of suicidal ideation. on the ASQ, suicidal ideation was more common than suicide attempts. The most common discharged disposition by the mental health team was to outpatient therapy. Use of medications and safety orders were not common within the cohort. Of the orders, searching patients for risk of self-harm and visual observations were the most common.

## Create an easy to re-use column grouping
bh_interven <- c(
  "ed_behavioral_health_screen_ind", "bhs_depression_score", "bhs_suicidal_ideation_score", "asq_suicidal_ideation_ind", "asq_suicide_attempt_ind",
  "mental_health_note_ind", "psych_tech_note_ind", "social_work_disposition_ind", "bhip_disposition_ind", "bhip_consult_ind", "medically_clear_ind", #' days_until_medically_clear',
  "bh_discharge_disposition"
)

med_interven <- c("behavioral_health_medication_given_ind", "agitation_order_set_ind", "ed_bh_com_order_set_ind", "edecu_bh_order_set_ind", "lorazepam_ind", "haloperidol_ind", "diphenhydramine_ind", "olanzapine_ind", "class_antianxiety_ind", "class_antihistamine_ind", "class_antipsychotic_ind")

safety_interven <- c("safety_risk_orders_ind", "restraints_ind", "care_suicidal_patient_proc_ind", "search_patients_risk_self_harm_ind", "visual_and_arms_length_ind", "visual_observation_ind", "ip_suicide_teaching_ind", "ed_bh_pathway_ind")

df_bh_interven_summary <- df %>%
  select(all_of(bh_interven), all_of(med_interven), all_of(safety_interven), total_ed_hrs)

## Create good looking summary tables
tbl7 <-
  df_bh_interven_summary %>%
  mutate_at(
    vars(ends_with("ind")),
    ~ (recode(., "1" = "Yes", "0" = "No"))
  ) %>%
  tbl_summary(
    # by = 'ED LOS Hrs Cat',
    include = c(all_of(bh_interven), all_of(med_interven), all_of(safety_interven)),
    statistic = list(all_continuous() ~ "{mean} ({sd})"),
    digits = all_continuous() ~ 2,
    sort = list(everything() ~ "frequency"),
    missing = "no"
  ) %>%
  add_n() %>%
  modify_caption("**Behavioral Health Activities**") %>%
  modify_header(label ~ "**Variable**")


tbl8 <-
  df_bh_interven_summary %>%
  mutate_at(
    vars(ends_with("ind")),
    ~ (recode(., "1" = "Yes", "0" = "No"))
  ) %>%
  tbl_continuous(
    variable = total_ed_hrs,
    include = c(all_of(bh_interven), all_of(med_interven), all_of(safety_interven))
  ) %>%
  modify_header(all_stat_cols() ~ "**Median LOS in Hours**")

tbl_final4 <-
  tbl_merge(list(tbl7, tbl8)) %>%
  modify_spanning_header(everything() ~ NA)

tbl_final4
Behavioral Health Activities
Variable N N = 7,9191 Median LOS in Hours2
ED Behavioral Health Screen 7,919 2,541 (32%)
No 6 (4, 9)
Yes 6 (4, 9)
BHS Depression Score 7,919
(Not Assessed) 5,378 (68%) 6 (4, 9)
Severe Depression 2,042 (26%) 6 (4, 9)
Mild Depression 341 (4.3%) 5 (4, 7)
Moderate Depression 158 (2.0%) 6 (4, 8)
BHS Suicidal Ideation Score 7,919
(Not Assessed) 7,578 (96%) 6 (4, 9)
No past or current SI 225 (2.8%) 5 (4, 7)
Lifetime SI 86 (1.1%) 6 (4, 8)
Current SI 30 (0.4%) 6 (4, 14)
ASQ Suicidal Ideation 7,919 1,837 (23%)
No 5 (4, 8)
Yes 7 (4, 30)
ASQ Suicide Attempt 7,919 1,091 (14%)
No 5 (4, 8)
Yes 8 (5, 40)
Mental Health Note 7,919 5,387 (68%)
No 5 (3, 7)
Yes 6 (4, 11)
Psych Tech Note 7,919 2,711 (34%)
No 5 (3, 7)
Yes 9 (6, 32)
Social Work Disposition 7,919 3,010 (38%)
No 6 (4, 8)
Yes 5 (4, 12)
BHIP Disposition 7,919 4,114 (52%)
No 5 (4, 7)
Yes 6 (4, 15)
BHIP Consult 7,919 3,093 (39%)
No 5 (3, 7)
Yes 8 (5, 17)
Medically Clear 7,919 1,670 (21%)
No 5 (4, 8)
Yes 7 (5, 33)
BH Discharge Disposition 3,145
Outpatient Therapy 1,317 (42%) 4 (3, 6)
Inpatient Hospitalization 977 (31%) 24 (8, 57)
Partial Hospitalization 767 (24%) 5 (4, 8)
No Intervention 51 (1.6%) 4 (3, 6)
Residential Treatment Facility 33 (1.0%) 6 (5, 9)
Unknown 4,774
Behavioral Health Medication Given 7,919 495 (6.3%)
No 5 (4, 8)
Yes 8 (6, 39)
Agitation Order Set 7,919 345 (4.4%)
No 6 (4, 9)
Yes 7 (6, 10)
ED BH Com Order Set 7,919 55 (0.7%)
No 6 (4, 9)
Yes 7 (5, 17)
EDECU BH Order Set 7,919 118 (1.5%)
No 6 (4, 8)
Yes 113 (52, 192)
Lorazepam 7,919 311 (3.9%)
No 6 (4, 8)
Yes 7 (6, 13)
Haloperidol 7,919 67 (0.8%)
No 6 (4, 9)
Yes 7 (6, 11)
Diphenhydramine 7,919 239 (3.0%)
No 6 (4, 8)
Yes 9 (6, 52)
Olanzapine 7,919 22 (0.3%)
No 6 (4, 9)
Yes 7 (5, 10)
Class Antianxiety 7,919 324 (4.1%)
No 6 (4, 8)
Yes 7 (6, 13)
Class Antihistamine 7,919 239 (3.0%)
No 6 (4, 8)
Yes 9 (6, 52)
Class Antipsychotic 7,919 96 (1.2%)
No 6 (4, 9)
Yes 7 (6, 11)
Safety Risk Orders 7,919 4,403 (56%)
No 5 (3, 7)
Yes 7 (5, 15)
Restraints 7,919 312 (3.9%)
No 6 (4, 8)
Yes 7 (6, 16)
Care Suicidal Patient Proc 7,919 1,167 (15%)
No 5 (4, 8)
Yes 10 (6, 52)
Search Patients Risk Self Harm 7,919 1,961 (25%)
No 5 (4, 7)
Yes 13 (6, 51)
Visual And Arms Length 7,919 1,183 (15%)
No 5 (4, 8)
Yes 7 (5, 11)
Visual Observation 7,919 4,021 (51%)
No 5 (3, 7)
Yes 7 (5, 18)
IP Suicide Teaching 7,919 375 (4.7%)
No 6 (4, 9)
Yes 7 (5, 11)
ED BH Pathway 7,919 2,005 (25%)
No 5 (4, 8)
Yes 7 (4, 25)

1 n (%)

2 Median (IQR)

Predictors of Length of Stays

Selecting Predictors

All Predictors

The following chart displays our predictors in a graphical form to help identify which predictors may be helpful in predictive statistical analyses. They are sorted based on contribution to the variation in LOS (R-squared).

df_pred_los <- df %>%
  select(all_of(pat_chars), all_of(dx_chars), all_of(ed_chars), all_of(bh_interven), all_of(med_interven), all_of(safety_interven), total_ed_hrs, -ed_los, -ed_los_hrs_cat, -primary_diagnosis_category) %>%
  rename_all(nice_display_names) %>% 
  rename(total_ed_hrs = "Total ED (Hours)")

plot_spread(df_pred_los, dep_var =  "total_ed_hrs")

Selected Predictors

Based on the strong relationship between the disposition variables and their temporal location to the discharge date (thus LOS), they were removed from further analyses.

ed_dispo_vars <- c('bh_discharge_disposition', 'ed_discharge_disposition', 'ed_psych_discharge_location', 'ed_discharge_location', 'social_work_disposition_ind', 'bhip_disposition_ind') 

analysis_vars <- df %>% 
select(all_of(pat_chars), all_of(dx_chars), all_of(ed_chars), all_of(bh_interven), all_of(med_interven), all_of(safety_interven), total_ed_hrs, -ed_los, -ed_los_hrs_cat, -primary_diagnosis_category, -all_of(ed_dispo_vars)) %>% 
  colnames(.)


df_pred_los_no_dispos <- df %>% 
    select(all_of(analysis_vars)) %>%
  rename_all(nice_display_names) %>% 
  rename(total_ed_hrs = "Total ED (Hours)")


plot_spread(df_pred_los_no_dispos, dep_var =  "total_ed_hrs")

Regression Analyses

For a given predictor variable, the coefficient (Beta) can be interpreted as the average effect on y of a one-unit increase in the predictor (LOS), holding all other predictors fixed.

Significant values are in bold. Variables with multiple levels are in comparison to the first value (Reference –). Thus, the values after it are in comparison to that Reference.

For example, a positive beta value would mean that there is an increase in LOS hours of the x LOS hours ( beta value) for that level in comparison to the Reference level.

Univariate Regression

The influence of each variable on its own. This is similar to the graph above but allows for the examination of different levels within variables.

Key Findings: Total Length of Stay was greater for

  • Patients over 9 compared to Patients under 9

  • Female patients compared to Male

  • Non-Hispanic black compared to Non-Hispanic White

  • Medical Assistance compared to Commercial insurance

  • Mental Health Diagnosis compared to Medical Diagnosis

  • Severe Depression compared to Mild Depression

  • Medication ordered compared to Not ordered

  • Safety procedures ordered compared to Not ordered

lm_vars <- analysis_vars

df_pat_lm <- df %>%
  select(all_of(lm_vars), total_ed_hrs)


df_pat_lm %>%
  mutate_at(
    vars(ends_with("ind")),
    ~ (recode(., "1" = "Yes", "0" = "No"))
  ) %>%
  tbl_uvregression(
    method = lm,
    y = (total_ed_hrs),
    show_single_row = vars(ends_with("ind")),
    pvalue_fun = ~ style_pvalue(.x, digits = 2)
  ) %>%
  bold_p() %>% # bold p-values under a given threshold (default 0.05)
  # bold_p(t = 0.10, q = TRUE) %>% # now bold q-values under the threshold of 0.10
  bold_labels() %>%
  modify_caption("**Univariate Regression: Predictors of Length of Stay**") %>%
  modify_header(label ~ "**Variable**") %>%
  italicize_levels()
Univariate Regression: Predictors of Length of Stay
Variable N Beta 95% CI1 p-value
Age Group 7917
5-8
9-12 12 8.5, 15 <0.001
13-15 8.1 5.2, 11 <0.001
16-18 5.0 2.0, 7.9 <0.001
Sex 7917
Male
Female 2.7 1.0, 4.4 0.002
Unknown -5.7 -78, 67 0.88
Race Ethnicity 7916
Non-Hispanic White
Asian 0.44 -4.5, 5.3 0.86
Hispanic or Latino -0.11 -3.1, 2.9 0.95
Non-Hispanic Black 2.0 0.17, 3.8 0.033
Other 0.62 -2.5, 3.8 0.70
County 7915
Philadelphia
Bucks -0.69 -4.3, 2.9 0.71
Chester -3.8 -8.5, 0.86 0.11
Delaware 2.9 0.40, 5.3 0.023
Montgomery 0.89 -1.9, 3.6 0.53
Other -4.6 -6.9, -2.3 <0.001
Preferred Language 7899
English
Spanish -2.5 -8.7, 3.8 0.44
Other -0.56 -7.5, 6.4 0.87
Payor Group 7796
Medical Assistance
Charity Care -15 -88, 58 0.69
Commercial -3.3 -5.0, -1.7 <0.001
Other -4.2 -10, 1.9 0.18
Mychop Activation 7722 1.7 -0.13, 3.5 0.069
Primary Diagnosis Is BH 7917 9.6 7.9, 11 <0.001
Primary BH Diagnosis Name 7917
(Medical diagnosis)
Aggression, agitation, anger 11 8.5, 13 <0.001
Altered mental status -3.9 -7.6, -0.16 0.041
Anxiety disorders -4.7 -8.4, -1.0 0.013
Eating disorders -3.8 -6.6, -1.0 0.007
Hallucinations -0.43 -9.5, 8.6 0.93
Other mental health condition 17 -2.0, 37 0.079
Somatic disorders -4.4 -14, 5.2 0.36
Substances use disorders 4.9 1.6, 8.2 0.004
Suicidal ideation, self-injury 29 27, 31 <0.001
Primary Medical Diagnosis Name 7917
(BH diagnosis)
Abdominal And Pelvic Pain -10 -16, -4.7 <0.001
Convulsions, Not Elsewhere Classified -12 -19, -4.0 0.003
Nausea And Vomiting -10 -18, -2.1 0.014
Overdose 2.1 -3.4, 7.7 0.45
Pain In Throat And Chest -13 -20, -4.9 0.001
Other -8.4 -10, -6.6 <0.001
Complex Chronic Condition 7917 -6.9 -8.8, -5.0 <0.001
Medically Complex 7917 -8.6 -11, -5.9 <0.001
Arrival Shift 7917
Day (8a-4p)
Evening (4p-12a) 1.2 -0.55, 2.9 0.18
Overnight (12a-8a) -0.69 -3.7, 2.3 0.65
Week Day 7917
Mon
Tue 2.2 -0.66, 5.0 0.13
Wed 2.3 -0.57, 5.1 0.12
Thu 1.7 -1.2, 4.5 0.26
Fri 2.0 -1.0, 4.9 0.19
Sat 3.1 -0.18, 6.4 0.064
Sun 0.37 -2.9, 3.6 0.82
School Month 7917 2.4 0.13, 4.7 0.039
ED Arrival Mode 7917
Car
Ems-other -3.7 -12, 4.2 0.36
Not Applicable 0.85 -1.0, 2.7 0.37
Private Vehicle -7.1 -9.4, -4.7 <0.001
Other -0.80 -4.4, 2.8 0.66
Psych Emergency Cc 7917 17 16, 19 <0.001
Altered Mental Status Cc 7917 -8.8 -14, -3.1 0.002
Mental Health Cc 7917 15 14, 17 <0.001
Hr 72 Revisit 7917 -2.7 -8.3, 2.8 0.33
ED Discharge Destination 7917
Patient/family/home
Not Applicable 5.8 3.5, 8.1 <0.001
Transfer To Another Facility 11 3.0, 20 0.008
Transfer To Psych Facility 15 9.7, 20 <0.001
Transfer To Rehab-cssh -4.6 -18, 9.0 0.50
Other -0.28 -10, 9.8 0.96
ED Behavioral Health Screen 7917 0.63 -1.1, 2.4 0.48
BHS Depression Score 7917
Mild Depression
Moderate Depression 5.6 -1.4, 13 0.12
Severe Depression 8.3 4.1, 13 <0.001
(Not Assessed) 6.4 2.4, 10 0.002
BHS Suicidal Ideation Score 7917
No past or current SI
Current SI 8.3 -5.8, 22 0.25
Lifetime SI 5.4 -3.8, 15 0.25
(Not Assessed) 9.0 4.1, 14 <0.001
ASQ Suicidal Ideation 7917 20 19, 22 <0.001
ASQ Suicide Attempt 7917 24 22, 27 <0.001
Mental Health Note 7917 14 12, 16 <0.001
Psych Tech Note 7917 28 27, 30 <0.001
BHIP Consult 7917 18 16, 19 <0.001
Medically Clear 7917 25 23, 27 <0.001
Behavioral Health Medication Given 7917 30 26, 33 <0.001
Agitation Order Set 7917 0.19 -3.8, 4.2 0.92
ED BH Com Order Set 7917 11 1.6, 21 0.022
EDECU BH Order Set 7917 122 116, 128 <0.001
Lorazepam 7917 14 9.5, 18 <0.001
Haloperidol 7917 5.5 -3.4, 14 0.22
Diphenhydramine 7917 39 34, 44 <0.001
Olanzapine 7917 7.8 -7.7, 23 0.33
Class Antianxiety 7917 13 9.2, 17 <0.001
Class Antihistamine 7917 39 34, 44 <0.001
Class Antipsychotic 7917 2.7 -4.7, 10 0.47
Safety Risk Orders 7917 18 16, 19 <0.001
Restraints 7917 7.2 3.0, 11 <0.001
Care Suicidal Patient Proc 7917 32 30, 35 <0.001
Search Patients Risk Self Harm 7917 36 35, 38 <0.001
Visual And Arms Length 7917 8.8 6.5, 11 <0.001
Visual Observation 7917 19 17, 21 <0.001
IP Suicide Teaching 7917 4.3 0.43, 8.1 0.029
ED BH Pathway 7917 18 17, 20 <0.001

1 CI = Confidence Interval

Multiple Regression

Analyzing the influence of all of the variables together on LOS.

Key Findings: Total Length of Stay was greater for

  • Patients 9-12 compared to Patients 5-8

  • Non-Hispanic black compared to Non-Hispanic White

  • Bucks, Delaware, Montgomery County compared to Philadelphia

  • Medical Assistance compared to Commercial Insurance

  • Mental Health Diagnosis compared to Medical Diagnosis

  • Not ordering safety procedures compared to Ordering

  • Ordering specific medications compared to Not ordering

lm_full_model <- lm(total_ed_hrs ~ ., data = df_pat_lm)

lm_full_model %>%
  tbl_regression(
    show_single_row= vars(ends_with("ind")),
    pvalue_fun = ~style_pvalue(.x, digits = 2)
  ) %>% 
  bold_p(t = 0.10) %>%
  bold_labels() %>%
  italicize_levels() %>% 
  modify_caption("**Multiple Regression: Predictors of Length of Stay**")  %>%
  modify_header(label ~ "**Variable**") %>% 
  add_glance_source_note()
Multiple Regression: Predictors of Length of Stay
Variable Beta 95% CI1 p-value
Age Group
5-8
9-12 4.0 1.3, 6.6 0.003
13-15 -1.5 -4.1, 1.2 0.28
16-18 -1.4 -4.1, 1.4 0.34
Sex
Male
Female 0.88 -0.62, 2.4 0.25
Unknown -28 -87, 31 0.35
Race Ethnicity
Non-Hispanic White
Asian 1.6 -2.6, 5.7 0.45
Hispanic or Latino -0.89 -3.7, 1.9 0.53
Non-Hispanic Black 1.7 -0.16, 3.6 0.072
Other 2.0 -0.70, 4.7 0.15
County
Philadelphia
Bucks 3.5 0.33, 6.7 0.031
Chester 0.10 -3.9, 4.1 0.96
Delaware 2.3 0.22, 4.4 0.030
Montgomery 3.8 1.3, 6.2 0.003
Other 2.2 -0.06, 4.4 0.056
Preferred Language
English
Spanish 3.1 -2.8, 9.0 0.30
Other 2.1 -4.2, 8.4 0.51
Payor Group
Medical Assistance
Charity Care 3.5 -55, 62 0.91
Commercial -1.8 -3.4, -0.14 0.033
Other -4.8 -9.8, 0.13 0.056
Mychop Activation -0.45 -2.0, 1.1 0.57
Primary Diagnosis Is BH 7.2 4.3, 10 <0.001
Primary BH Diagnosis Name
(Medical diagnosis)
Aggression, agitation, anger -2.2 -5.1, 0.72 0.14
Altered mental status -6.5 -11, -2.3 0.003
Anxiety disorders -7.2 -11, -3.3 <0.001
Eating disorders -17 -20, -13 <0.001
Hallucinations -11 -19, -2.6 0.010
Other mental health condition 3.8 -13, 21 0.66
Somatic disorders -7.5 -16, 1.1 0.086
Substances use disorders -5.5 -8.9, -2.1 0.002
Suicidal ideation, self-injury
Primary Medical Diagnosis Name
(BH diagnosis)
Abdominal And Pelvic Pain 2.6 -2.7, 7.9 0.34
Convulsions, Not Elsewhere Classified 1.8 -4.9, 8.6 0.60
Nausea And Vomiting 1.7 -5.3, 8.8 0.63
Overdose -9.0 -14, -3.7 <0.001
Pain In Throat And Chest 0.18 -6.6, 7.0 0.96
Other 0.70 -2.0, 3.4 0.61
Complex Chronic Condition -0.49 -2.6, 1.6 0.64
Medically Complex 1.8 -1.0, 4.6 0.21
Arrival Shift
Day (8a-4p)
Evening (4p-12a) 0.08 -1.4, 1.5 0.92
Overnight (12a-8a) -1.8 -4.4, 0.68 0.15
Week Day
Mon
Tue 1.3 -1.0, 3.6 0.28
Wed 3.3 0.94, 5.6 0.006
Thu 1.4 -0.93, 3.8 0.24
Fri 2.4 -0.08, 4.8 0.058
Sat 3.4 0.66, 6.1 0.015
Sun 1.9 -0.73, 4.6 0.16
School Month 1.3 -0.55, 3.2 0.16
ED Arrival Mode
Car
Ems-other -3.4 -10, 3.3 0.32
Not Applicable 0.55 -1.0, 2.1 0.50
Private Vehicle -1.7 -3.7, 0.38 0.11
Other 1.6 -1.4, 4.6 0.28
Psych Emergency Cc 4.4 1.5, 7.3 0.003
Altered Mental Status Cc -0.88 -5.9, 4.1 0.73
Mental Health Cc -3.9 -6.7, -1.1 0.007
Hr 72 Revisit 0.34 -4.2, 4.9 0.88
ED Discharge Destination
Patient/family/home
Not Applicable 3.2 1.1, 5.3 0.003
Transfer To Another Facility -1.0 -7.9, 6.0 0.79
Transfer To Psych Facility -3.8 -8.5, 0.86 0.11
Transfer To Rehab-cssh 2.3 -8.9, 13 0.69
Other 2.5 -5.9, 11 0.56
ED Behavioral Health Screen 1.2 -2.9, 5.3 0.57
BHS Depression Score
Mild Depression
Moderate Depression 3.5 -2.6, 9.6 0.26
Severe Depression 0.52 -3.7, 4.8 0.81
(Not Assessed)
BHS Suicidal Ideation Score
No past or current SI
Current SI -2.8 -14, 8.5 0.63
Lifetime SI -2.1 -9.5, 5.4 0.58
(Not Assessed)
ASQ Suicidal Ideation -0.66 -3.1, 1.8 0.60
ASQ Suicide Attempt 4.8 2.3, 7.3 <0.001
Mental Health Note -4.2 -6.2, -2.1 <0.001
Psych Tech Note 14 12, 16 <0.001
BHIP Consult 5.8 3.7, 8.0 <0.001
Medically Clear 5.1 3.2, 7.1 <0.001
Behavioral Health Medication Given -18 -34, -3.1 0.018
Agitation Order Set -0.58 -16, 15 0.94
ED BH Com Order Set 13 -1.4, 27 0.078
EDECU BH Order Set 90 75, 106 <0.001
Lorazepam 27 11, 44 0.001
Haloperidol 17 0.62, 34 0.042
Diphenhydramine 13 7.2, 19 <0.001
Olanzapine 22 3.7, 40 0.018
Class Antianxiety -20 -38, -3.3 0.019
Class Antihistamine
Class Antipsychotic -21 -37, -4.3 0.013
Safety Risk Orders -6.8 -11, -3.0 <0.001
Restraints -3.8 -8.2, 0.63 0.092
Care Suicidal Patient Proc -3.5 -6.9, -0.14 0.041
Search Patients Risk Self Harm 30 27, 33 <0.001
Visual And Arms Length -9.1 -12, -6.4 <0.001
Visual Observation 3.1 -0.42, 6.6 0.085
IP Suicide Teaching -16 -19, -12 <0.001
ED BH Pathway 3.2 1.2, 5.3 0.002
R² = 0.374; Adjusted R² = 0.366; Sigma = 29.5; Statistic = 50.9; p-value = <0.001; df = 88; Log-likelihood = -36,428; AIC = 73,035; BIC = 73,659; Deviance = 6,524,089; Residual df = 7,505; No. Obs. = 7,594

1 CI = Confidence Interval

Multiple Logistic Regression

Predicting prolonged LOS from the variables selected for analysis. Interpreting the Odds Ratio (OR) means that positive values indicate a higher likelihood that the variable has a higher chance of having a prolonged length of stay.

Findings:
Being a female patient and Non-Hispanic Blacks were associated with prolonged length of stays in comparison to male and Non-Hispanic White patients.

Having a primary diagnosis that was behavioral health was associated with prolonged length of stays. Among the behavioral health diagnoses, conditions such as anxiety, substance use, and aggression, agitation, and anger were associated with longer length of stays in comparison to medical diagnosis. This is consistent with the findings that having a chief complaint of psych emergency or a general mental health chief complaint was associated with prolonged length of stays.

Arriving at the ED in the evening and overnight were associated with prolonged length of stays in comparison to arrivals earlier in the day.

Among the safety orders available, having any of the safety orders except for restraints was associated with a longer length of stay.

Medications ordered did not have a significant association, nor did depression scores or suicidal ideation levels.

df_pat_glm <- df %>%
  select(all_of(lm_vars), ed_los_hrs_cat, -total_ed_hrs) %>%
  mutate(ed_los_hrs_cat = recode_factor(ed_los_hrs_cat, "ED LOS < 24" = 1, "ED LOS >= 24" = 0))


glm_full_model <- glm(ed_los_hrs_cat ~ .,
  data = df_pat_glm,
  family = binomial()
)

glm_full_model %>%
  tbl_regression(
    exponentiate = TRUE,
    show_single_row = vars(ends_with("ind")),
    pvalue_fun = ~ style_pvalue(.x, digits = 2)
  ) %>%
  bold_p(t = 0.10) %>%
  bold_labels() %>%
  italicize_levels() %>%
  modify_caption("**Logistic Regression: Predictors of Prolonged Length of Stay**") %>%
  modify_header(label ~ "**Variable**") %>%
  add_significance_stars(hide_ci = FALSE, hide_p = FALSE) 
Logistic Regression: Predictors of Prolonged Length of Stay
Variable OR1,2 SE2 95% CI2 p-value
Age Group
5-8
9-12 1.12 0.309 0.62, 2.09 0.71
13-15 0.91 0.310 0.50, 1.70 0.77
16-18 0.86 0.324 0.46, 1.66 0.65
Sex
Male
Female 1.39* 0.138 1.06, 1.83 0.016
Unknown 0.00 10,754 >0.99
Race Ethnicity
Non-Hispanic White
Asian 1.63 0.373 0.77, 3.34 0.19
Hispanic or Latino 1.18 0.245 0.73, 1.89 0.51
Non-Hispanic Black 1.43* 0.161 1.05, 1.97 0.026
Other 0.99 0.242 0.62, 1.59 0.98
County
Philadelphia
Bucks 1.42 0.276 0.82, 2.43 0.21
Chester 1.83 0.371 0.87, 3.74 0.10
Delaware 1.07 0.170 0.76, 1.49 0.70
Montgomery 1.55* 0.208 1.03, 2.33 0.035
Other 0.78 0.236 0.49, 1.23 0.29
Preferred Language
English
Spanish 0.96 0.556 0.31, 2.75 0.94
Other 2.81 0.531 0.95, 7.64 0.052
Payor Group
Medical Assistance
Charity Care 0.00 10,754 >0.99
Commercial 0.73* 0.148 0.54, 0.97 0.030
Other 0.32* 0.482 0.12, 0.80 0.020
Mychop Activation 1.15 0.130 0.89, 1.49 0.27
Primary Diagnosis Is BH 2.49*** 0.234 1.59, 3.98 <0.001
Primary BH Diagnosis Name
(Medical diagnosis)
Aggression, agitation, anger 0.73 0.197 0.50, 1.08 0.12
Altered mental status 0.14 1.05 0.01, 0.73 0.063
Anxiety disorders 0.31* 0.600 0.08, 0.91 0.049
Eating disorders 0.00 323 0.00, 0.00 0.96
Hallucinations 0.30 0.757 0.06, 1.19 0.11
Other mental health condition 0.41 2.03 0.01, 8.62 0.66
Somatic disorders 0.00 1,422 0.00, 0.00 >0.99
Substances use disorders 0.51** 0.247 0.31, 0.83 0.007
Suicidal ideation, self-injury
Primary Medical Diagnosis Name
(BH diagnosis)
Abdominal And Pelvic Pain 0.95 1.08 0.05, 5.25 0.96
Convulsions, Not Elsewhere Classified 1.23 1.09 0.06, 7.12 0.85
Nausea And Vomiting 0.00 1,119 0.00, 0.00 0.99
Overdose 1.00 0.344 0.51, 1.96 >0.99
Pain In Throat And Chest 0.00 1,099 0.00, 0.00 0.99
Other 1.48 0.267 0.88, 2.51 0.14
Complex Chronic Condition 0.62* 0.205 0.41, 0.92 0.020
Medically Complex 0.79 0.361 0.38, 1.57 0.50
Arrival Shift
Day (8a-4p)
Evening (4p-12a) 0.74* 0.127 0.58, 0.95 0.020
Overnight (12a-8a) 0.72 0.210 0.48, 1.09 0.12
Week Day
Mon
Tue 0.92 0.204 0.62, 1.37 0.67
Wed 1.00 0.208 0.67, 1.51 0.99
Thu 0.68 0.210 0.45, 1.02 0.065
Fri 0.72 0.221 0.47, 1.12 0.15
Sat 0.97 0.236 0.61, 1.54 0.89
Sun 1.44 0.233 0.91, 2.27 0.12
School Month 1.11 0.174 0.79, 1.56 0.56
ED Arrival Mode
Car
Ems-other 0.56 0.517 0.19, 1.45 0.26
Not Applicable 0.75* 0.133 0.58, 0.97 0.029
Private Vehicle 0.60* 0.227 0.38, 0.93 0.025
Other 0.85 0.254 0.51, 1.39 0.52
Psych Emergency Cc 1.87* 0.285 1.08, 3.29 0.028
Altered Mental Status Cc 0.18 1.11 0.01, 1.08 0.12
Mental Health Cc 0.47* 0.332 0.24, 0.89 0.023
Hr 72 Revisit 1.24 0.466 0.48, 2.96 0.64
ED Discharge Destination
Patient/family/home
Not Applicable 2.27*** 0.210 1.51, 3.45 <0.001
Transfer To Another Facility 0.90 0.478 0.34, 2.23 0.83
Transfer To Psych Facility 0.81 0.298 0.45, 1.45 0.48
Transfer To Rehab-cssh 12.4* 1.20 0.56, 89.0 0.036
Other 4.78* 0.665 1.18, 16.5 0.019
ED Behavioral Health Screen 1.10 0.706 0.23, 3.74 0.90
BHS Depression Score
Mild Depression
Moderate Depression 1.08 0.844 0.22, 6.37 0.93
Severe Depression 1.04 0.710 0.30, 4.98 0.96
(Not Assessed)
BHS Suicidal Ideation Score
No past or current SI
Current SI 1.88 1.00 0.28, 14.6 0.53
Lifetime SI 0.22 0.98 0.03, 1.60 0.12
(Not Assessed)
ASQ Suicidal Ideation 1.28 0.173 0.91, 1.80 0.15
ASQ Suicide Attempt 1.21 0.157 0.89, 1.65 0.22
Mental Health Note 1.27 0.302 0.71, 2.34 0.43
Psych Tech Note 17.5*** 0.237 11.1, 28.2 <0.001
BHIP Consult 1.43* 0.168 1.03, 1.99 0.032
Medically Clear 1.23 0.130 0.95, 1.59 0.11
Behavioral Health Medication Given 0.46 1.01 0.06, 3.15 0.44
Agitation Order Set 1.30 1.05 0.17, 10.6 0.80
ED BH Com Order Set 1.17 0.930 0.18, 6.93 0.87
EDECU BH Order Set 26.0** 1.05 3.67, 227 0.002
Lorazepam 5.19 1.48 0.39, 83.6 0.26
Haloperidol 0.23 1.04 0.03, 2.00 0.16
Diphenhydramine 0.68 0.420 0.30, 1.54 0.36
Olanzapine 0.53 1.23 0.05, 6.07 0.60
Class Antianxiety 0.25 1.49 0.02, 3.61 0.36
Class Antihistamine
Class Antipsychotic 3.95 0.98 0.51, 26.0 0.16
Safety Risk Orders 0.20** 0.496 0.07, 0.52 0.001
Restraints 1.23 0.285 0.70, 2.14 0.47
Care Suicidal Patient Proc 0.59** 0.175 0.42, 0.82 0.002
Search Patients Risk Self Harm 27.0*** 0.198 18.4, 40.1 <0.001
Visual And Arms Length 0.30*** 0.159 0.22, 0.40 <0.001
Visual Observation 2.91** 0.394 1.41, 6.74 0.007
IP Suicide Teaching 0.47*** 0.202 0.31, 0.69 <0.001
ED BH Pathway 1.28 0.129 0.99, 1.65 0.058

1 *p<0.05; **p<0.01; ***p<0.001

2 OR = Odds Ratio, SE = Standard Error, CI = Confidence Interval

library(sjPlot)

plot_model(glm_full_model, sort.est = TRUE, show.values = TRUE, value.offset = .4,  transform = NULL, type = "std") + theme_chop() 

Conclusions

Patient factors are related to length of stay, consistent with the literature: Age, Race Ethnicity, Sex, and Insurance .

Suggests further analysis for whether there are socioeconomic disparities and what ED practices may contribute to them.

ED factors and clinical activities had some influence as well Consultation with psychiatry and needing safety orders suggests likely higher acuity and intervention needed.

Next steps Consider using predictive modeling techniques Varying prolonged LOS cutoffs.

Limitations: Lots of missing data as not all factors are consistently measured. Limited the ability to do predictive analyses.

Supplemental Analyses

Extra analyses for my curiosity.

Census Count of ED Patients

Daily Census for the ED. This measures the number of patients who were in the ED on a given day. Includes new patients as well as those previously admitted but not yet discharged.

df_daily_ed_patients <- df %>% 
  select(census_date, visit_key) %>% 
  remove_incomplete_end_dates(census_date, period = 'month') %>% 
  add_date_columns(census_date) %>% 
  group_by(census_date_month) %>% 
  summarize(count = n())

  hc_spc(
     data = df_daily_ed_patients,
    x = census_date_month,
    y = count, 
    chart = "c",
    title = "Median Number of Patients Present in the ED", xlab = "Month Year", ylab = "Patients", agg.fun = "median"
  )

New Patients to the ED

Examining the total of new patients admitted to the ED during the period. Does not include patients who were already present in the ED.

df_admission_count <- df %>% 
  rocqi::remove_incomplete_end_dates(ed_admit_date, period = 'month') %>% 
  group_by(ed_admit_date) %>% 
  summarise(count = n()) 

hc_spc(
  data = df_admission_count, x = ed_admit_date,
  y = count, chart = "c",
  ylab = "Total Admissions"
)

Disposition Flow

This chart illustrates the location where the patient was discharged from and the disposition recommended by the BHIP and Social Work teams.

df_hc <- df %>% 
  #filter(df$encounter_date >= '2021-01-01') %>%
  select(ed_discharge_location, bh_discharge_disposition) %>%
  drop_na() %>% 
  data_to_sankey() %>%
  mutate(to = as.factor(to)) %>%
  mutate(to = fct_reorder(to, weight, .desc = TRUE) ) 
 
hchart((df_hc), "sankey", name = "Behavioral Health Patient Discharge Dispositions") %>%
  hc_title(text= "Behavioral Health Patient Discharge Dispositions") %>%
  hc_subtitle(text= "CHOP Unit to BH Disposition") %>% 
      hc_tooltip(pointFormat = "<b>Value:</b> {point.weight} <br>
                 <b>Percentage</b> {point.percentage:,.2f}%")

Placement for Prolonged LOS

Comparing the ED Length of Stay to the disposition recommendations by the behavioral health team.

df_hc_long_los <- df %>% 
  #filter(df$encounter_date >= '2021-01-01') %>%
  select(ed_los_hrs_cat, bh_discharge_disposition) %>%
  drop_na() %>% 
  data_to_sankey() %>%
  mutate(to = as.factor(to)) %>%
  mutate(to = fct_reorder(to, weight, .desc = TRUE) ) %>% 
  group_by(from) %>% 
  mutate(percentage = (weight/sum(weight))*100)
 
hchart((df_hc_long_los), "sankey", name = "Prolonged LOS Dispositions") %>%
  hc_title(text= "Behavioral Health Patient Discharge Dispositions") %>%
  #hc_subtitle(text= "CHOP Unit to BH Disposition") %>% 
      hc_tooltip(pointFormat = "<b>Value:</b> {point.weight} <br>
                 <b>Percentage</b> {point.percentage:,.2f}%")

Behavioral Health Diagnosis and Disposition Recommendations

This is the relationship between the primary ED diagnosis and the disposition recommendations by the behavioral health team.

df_hc_dx_dispo <- df %>% 
    filter(primary_diagnosis_is_bh_ind ==1) %>% 
  select(primary_bh_diagnosis_name, bh_discharge_disposition) %>%
  drop_na() %>% 
  data_to_sankey() %>%
  mutate(to = as.factor(to)) %>%
  mutate(to = fct_reorder(to, weight, .desc = TRUE) ) %>% 
  group_by(from) %>% 
  mutate(percentage = (weight/sum(weight))*100)
 
hchart((df_hc_dx_dispo), "sankey", name = "Prolonged LOS Dispositions") %>%
  hc_title(text= "Behavioral Health Diagnosis and Disposition Recommendations") %>%
  #hc_subtitle(text= "CHOP Unit to BH Disposition") %>% 
      hc_tooltip(pointFormat = "<b>Value:</b> {point.weight} <br>
                 <b>Percentage</b> {point.percentage:,.2f}%")

Transfer to Inpatient Psych Facility from ED

Inpatient Psychiatric Facility counts by the location where the patient was discharged from.

df_ip_psych_transfer <- df %>% 
  select(csn , ed_discharge_location, ed_psych_discharge_location) %>% 
  filter(ed_discharge_location %in% c('ED', 'EDECU')) %>% 
  group_by(ed_discharge_location, ed_psych_discharge_location) %>% 
  summarise(count = n()) %>% 
pivot_wider(names_from = ed_discharge_location, values_from = count) %>% 
  group_by(ed_psych_discharge_location) %>% 
  filter(!is.na(ed_psych_discharge_location)) %>% 
  mutate("Total" = sum(c(ED,EDECU), na.rm=TRUE)) %>% 
  rename("Psychiatric Facility" = ed_psych_discharge_location) %>% 
  format_data_frame(keep_uppercase = c("bh", "edecu", "bhs", "bhip", "mh", "ed", "csn", "mrn", "ip"))

df_ip_psych_transfer %>%
  kbl() %>%
    kable_styling(bootstrap_options = c("striped", "hover", "condensed", "responsive"))
Psychiatric Facility ED EDECU Total
Belmont 11 100 111
Devereux 1 8 9
Fairmount 4 159 163
Foundations 6 16 22
Friends 8 46 54
Horsham Clinic 4 112 116
Kids Clinic NA 11 11
Rockford NA 7 7

Appendix

Inclusion Variables

  • Mental Health Note: Patient had at least one note (MH-type) documented during the admission by the behavioral health inpatient (BHIP) consult team.

  • Psych Tech Note: Patient had at least one note documented by a psychiatry tech during the admission.

  • Safety Risk Orders: Patient had at least one of the following safety risk orders during their admission.

    • Restraints

    • Visual & Arms Length

    • Visual Observation Care of the Suicidal

    • Patient Searching Patients at Risk for Self-Harm

    • IP Suicide Teaching

    • ED Pathway

  • Behavioral Health Behavioral Health Screen - ED: Patient has been screened using the behavioral health screen in the ED and obtained either a depression score of ‘Severe Depression’ or an ideation score of ‘Current Suicidal Thoughts.’

  • Social Work Disposition: Patient has a discharge disposition documented during the admission from the social work team.

  • BHIP Disposition: Patient has a discharge disposition documented during the admission from the BHIP team.

  • Medical Clearance: Patient has a medical clearance date documented during their admission.

  • Primary Diagnosis: Patient’s primary diagnosis during admission was of an ICD-10 diagnosis.

  • Medication Intervention: Patient any of the following Order Sets placed during their admission.

    • Agitation Medications

    • ED Behavioral Health Complaint Pathway

    • EDECU Behavioral Health Order Set

  • Medically Complex: Indicates whether a patient is listed as medically complex, which is either two complex chronic conditions, or one CCC and is reported as technology dependent.

  • Complex Chronic Condition: Indicates whether a patient has a complex chronic condition (CCC) based on specific diagnoses on their problem list, or if they have a specific visit diagnosis made in the past year. CCCs are grouped in to the following categories based on organ group - Hematology, Renal, GI, Malignancy, Neonatal, Congenital Genetics, Respiratory, Cardiovascular Disease, and Neuromuscular.

Bibliography

Case, Sarah D., Brady G. Case, Mark Olfson, James G. Linakis, and Eugene M. Laska. 2011. “Length of Stay of Pediatric Mental Health Emergency Department Visits in the United States.” Journal of the American Academy of Child & Adolescent Psychiatry 50 (11): 1110–19. https://doi.org/10.1016/j.jaac.2011.08.011.
Chakravarthy, Bharath, Allen Yang, Uzor Ogbu, Carole Kim, Anum Iqbal, Joanna Haight, Craig Anderson, et al. 2017. “Determinants of Pediatric Psychiatry Length of Stay in 2 Urban Emergency Departments.” Pediatric Emergency Care 33 (9): 613–19. https://doi.org/10.1097/pec.0000000000000509.
Nash, Katherine A., Bonnie T. Zima, Craig Rothenberg, Jennifer Hoffmann, Claudia Moreno, Marjorie S. Rosenthal, and Arjun Venkatesh. 2021. “Prolonged Emergency Department Length of Stay for US Pediatric Mental Health Visits (20052015).” Pediatrics 147 (5): e2020030692. https://doi.org/10.1542/peds.2020-030692.
Whitney, Daniel G., and Mark D. Peterson. 2019. “US National and State-Level Prevalence of Mental Health Disorders and Disparities of Mental Health Care Use in Children.” JAMA Pediatrics 173 (4): 389. https://doi.org/10.1001/jamapediatrics.2018.5399.